Hybrid Machine Learning Approach For Electric Load Forecasting

Electricity not only plays a vital role in our daily lives, but it is also extremely important in national economic and social development. Accurate electric load forecasting can help electric power industry to secure electricity supply and use scheduling to reduce waste of electricity. In this paper, we propose a novel hybrid machine learning combining long short time memory and gradient descent approach to forecast the future hourly electricity demand of northern Taiwan. Furthermore, Random forest is applied to explore the influence of temperature of each city in northern Taiwan and features of electric load and remove irrelevant factors. The experimental results confirm the effectiveness of the proposed hybrid machine learning approach. The average error percentage of load forecast is less than 2.5%.